{ "background": "The operational efficiency of industrial machinery fleets is a critical determinant of productivity in developing economies. In Ghana, a lack of robust methodological frameworks for evaluating fleet performance hinders systematic yield optimisation in key sectors such as mining, construction, and agriculture.", "purpose and objectives": "This study conducts a comparative methodological evaluation of fleet management systems, with the primary objective of determining the most effective analytical approach for measuring and predicting yield improvements. It aims to identify key operational variables influencing output.", "methodology": "A comparative study was performed using operational data from multiple industrial sites. A multilevel regression model, $Y{ij} = \\beta{0j} + \\beta{1j}X{1ij} + \\beta{2}X{2ij} + r{ij}$, with $\\beta{0j} = \\gamma{00} + \\gamma{01}Z{j} + u{0j}$, was employed, where $i$ indexes machinery and $j$ indexes sites. Model comparisons were based on Akaike Information Criterion and robust standard errors.", "findings": "The multilevel model significantly outperformed traditional pooled regression, explaining 34% more variance in yield. A key finding was that preventive maintenance compliance had a stronger positive effect on yield (β = 0.42, 95% CI [0.31, 0.53]) than machine age or fuel type.", "conclusion": "The application of multilevel regression provides a superior methodological framework for analysing nested industrial fleet data, offering more accurate insights for yield optimisation compared to conventional single-level techniques.", "recommendations": "Industry practitioners should adopt hierarchical modelling techniques for fleet performance analysis. Policymakers should support the development of standardised data collection protocols to facilitate such advanced analyses across sectors.", "key words": "fleet management, multilevel modelling, regression analysis, operational efficiency, yield optimisation, industrial engineering", "contribution statement": "This paper provides a novel comparative validation of mult